Background
One of the most important factors in assessing the health risks posed by an epidemic of infectious disease is pathogen transmissibility. Widespread public anxiety during the 2013–2016 West African Ebola epidemic, while driven by an extremely high fatality rate during the early stages, was fueled in part by the speed with which Ebola Virus Disease (EVD) spread throughout the populations of Liberia, Sierra Leone, and Guinea [
1‐
3]. Similarly, concerns over annual influenza epidemics in the United States center on densely inhabited areas with multiple opportunities for viral transmission due to physical proximity between susceptible individuals [
4]. Regardless of geographic setting, understanding how to slow and control pathogen dissemination is a high priority in forecasting and preventing epidemics of infectious disease.
Epidemic modelers frequently employ compartmental models of disease outbreaks, such as Susceptible-Infected-Recovered (SIR) models as in Keeling and Rohani [
5], Susceptible-Infected-Susceptible (SIS) models as in Gray et al [
6], and Susceptible-Exposed-Infected-Recovered (SEIR) models as in LeGrand et al [
7]. Accurately estimating and modeling the number of infected and susceptible individuals in at-risk populations is of crucial importance in these models. Such estimation is complicated, however, by efforts to isolate infected individuals in healthcare facilities or other community settings to decrease contact with the susceptible population. While the isolation of infected individuals is beneficial and should be encouraged, it challenges data analysts because it is time-varying and reflects dynamic and often unpredictable human behavior. Moreover, the rate at which infected individuals are removed from the population typically accelerates throughout an epidemic as awareness of the infectious threat increases [
8], a process Drake et al referred to as “societal learning” [
9]. Obtaining accurate estimates of this time-varying removal of infected persons, while difficult, improves the quality of compartmental models for epidemics of infectious disease [
9,
10]. To our knowledge, however, no work has directly compared the rate of behavioral adaption across multiple epidemics, societies, and geographic settings.
Many factors can affect how quickly effective isolation practices are implemented, such as access to health care, local public health funding, international aid, and the efficacy of information campaigns [
11]. Local health care practices and non-formal healthcare systems also provide care to patients during epidemics and can play a part in quarantining infected individuals [
12]. Previous work in Liberia has shown that a combination of these approaches through simultaneous community engagement and clinical intervention is more effective than any single intervention, with both health care access and utilization increasing hand-in-hand to decrease EVD transmission during the 2013–2016 Ebola epidemic [
13]. While infection prevention and control practices often include vaccination, progress to develop effective vaccines for emerging infections is slow and not necessarily more effective than isolation of infected individuals [
14]. Ring vaccination with the rVSV-ZEBOV-GP Ebola vaccine [
15] in the Democratic Republic of the Congo is promising [
16], but previous work has suggested that ring vaccination may only provide a marginal benefit to rigorous contact tracing and patient isolation [
17].
This paper aims to identify key similarities and differences in the behavioral response to outbreaks of three emerging zoonotic infections, focusing particularly on the speed with which infected individuals are removed from the susceptible population. We sought to determine how the mean removal rate (the inverse of the number of days from symptom onset to isolation of infected individuals) changed over the course of each outbreak as measured by epidemic week and viral serial interval. Individuals often experience zoonotic and emerging infections as innately more frightening than “familiar” diseases, leading to rapid behavioral adaptations due to high perceived risk of severe disease and death [
18]. Behavior modification, while crucial for epidemic containment [
19‐
21], is context dependent and difficult to predict due to social network, socioeconomic, and behavioral differences between populations [
22]. Thus, we chose seven different outbreaks of disease that stoked significant local and international fear due to the risk of global pandemic: the 2013–2016 Liberian Ebola epidemic, subsets of the 2013–2016 Liberian outbreak from Lofa and Montserrado Counties [
23], the 2003 Hong Kong SARS epidemic [
24‐
26], the 2014 Saudi Arabia MERS outbreaks in Riyadh and Jeddah [
27], and the 2015 South Korea MERS outbreak [
28]. We examined whether epidemic week and serial interval successfully predicted days from disease onset to hospitalization (DSOH) and mean removal rate (MRR) throughout each epidemic.
Discussion
The models presented in this study show that the removal of infected individuals from the susceptible population increases over time and varies significantly based on outbreak duration and location. Our findings suggest that populations perceive the health risks posed by emerging and re-emerging pathogens and respond rapidly to limit community transmission. Such population-level quarantine behaviors develop quickly in extremely disparate settings (Liberia, Hong Kong, Saudi Arabia, and South Korea) and distinct viral outbreaks (SARS, MERS, and Ebola). While DSOH improved (decreased) in every epidemic over time, extreme disparities in starting values (approximately 13 days from symptom onset to hospitalization at the beginning of the 2013–2016 Ebola outbreak in Liberia, versus approximately 5 days in the 2015 MERS outbreak in South Korea) highlight the intrinsic disadvantage that low-income countries may experience due to the interrelated concerns of poverty, limited access to health care, and low investment in public health. DSOH and MRR regressed against epidemic week differed across all observed outbreaks, and MRR likewise differed markedly based on the virus in question (Ebola, MERS, or SARS), the location, and at times both. Both DSOH and MRR are useful measurements of public health behavior during outbreaks, and are useful tools to compare outbreak response effectiveness in distinct geographic, economic, and social settings. Of course, DSOH and MRR are intrinsically and simply related since one is simply the reciprocal of the other. The main advantage of DSOH is that it is expressed in intuitive units (days elapsed), whereas MRR reflects the theoretical “removal rate” of standard compartmental models [
5].
Figure
2 highlights differences in mean change of the MRR due to outbreak type (Ebola, MERS, or SARS) and location. Mean change of the MRR was similar when calculated using epidemic week versus serial interval for Ebola, but demonstrated a lower estimate and lower standard error when calculated using serial interval in all three outbreaks of MERS and the outbreak of SARS in Hong Kong. This suggests that the relevance of various predictors (epidemic week versus serial interval) may vary based upon the type and location of an outbreak, although the comparative relevance of epidemic type versus location cannot be disentangled with the data available in this study. We recommend similar analyses of MRR be conducted across a wide range of geographies as outbreaks of emerging pathogens arise, providing important data on the range of MRR, and its expected rate of change, in different settings. This approach would be further strengthened by incorporating data regarding other context- and culture-specific social conditions, such as the presence of domestic and international aid workers, access to health services, and density and frequency of social interaction. Comparing findings between different viral epidemics is also challenged by different periods and mechanisms of viral transmission (e.g. respiratory droplets versus aerosols, presence or absence of transmissibility before symptoms, etc.).
While our findings demonstrate large and statistically significant differences in MRR, it is notable that the calculated rates of change in the MRRs are within a factor of ten (when calculated using epidemic week) to seven (when calculated using serial interval) of each other (Fig.
2), with the mean change being the lowest in the EVD outbreak in Liberia and the highest in the MERS outbreak in South Korea. For modelers seeking to understand the epidemiology of emerging infectious diseases with limited or no data from previous outbreaks, this study provides a range of acceptable values for the MRR based on seven geographically distinct outbreaks of three emerging diseases. Similarly, while large disparities in DSOH are obvious (Fig.
1), these data highlight that all societies quickly adapt to outbreaks of emerging infections. Drake et al previously demonstrated the positive impact of behavior change in infectious outbreaks, noting that doubling the rate of “societal learning” in a model of the 2003 SARS outbreak in Singapore approximately halved the estimated number of infected patients [
9]. While there is a theoretical upper limit to the speed with which newly-infected individuals can be removed from the susceptible population [
9], public health strategies aimed at fostering behavioral adaptations and accelerating isolation should form a cornerstone of interventions tasked with limiting the spread of highly contagious and deadly emerging pathogens [
31]. Such strategies extend beyond what we have presented here, such as prompt isolation of infected individuals within hospitals and other health care settings, and include other collective measures such as household quarantine practices, remote work, cancellation of mass gatherings, limitations on travel, use of facemasks or other virus-specific barriers against transmission, immunizations (when available), and more.
Conclusions
We have shown that rates of hospital or healthcare facility isolation of infected individuals from the susceptible population vary significantly by pathogen and location, but can in some cases be predicted by the timing and serial interval of the epidemic. This study detected variation in DSOH and MRR based on epidemic location and outbreak type, indicating that it may be possible to estimate a general range of the rate of change in these variables over time. Due to location-specific differences in DSOH and MRR, modelers who seek to develop forecasts early in an outbreak would benefit from estimating an expected range for removal of infected individuals using data from past outbreaks of the same pathogen in a similar setting. Furthermore, the quality of these estimates will be impacted by the metric chosen, as seen by the notable, but distinct, trends detected in DSOH and MRR. As seen in this study, even a relatively small amount of data from an emerging or re-emerging outbreak of infectious disease can be paired with a well-chosen response variable such as DSOH or MRR to make effective forecasts about public health behavior.
Open AccessThis article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit
http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (
http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.